Multi-parameter fusion motor performance evaluation method

Abstract

The invention relates to a multi-parameter fusion motor performance evaluation method, equipment and a medium. The method comprises the following steps: firstly, acquiring electrical signals, mechanical vibration signals and thermal temperature data in operation of a motor, and performing preprocessing and standardization to form a first data set; secondly, dimension reduction is carried out on multi-dimensional parameters of the first data set, and core parameter combinations are extracted to generate a second feature data set; then, classifying and marking the second feature data set by utilizing a classification algorithm, and determining a key performance index distribution interval to obtain a third classification result set; and finally, constructing a time sequence anomaly detection model based on the third classification result set, analyzing an index change trend, identifying an abnormal point deviating from a preset threshold value, and generating a fourth anomaly detection result. According to the method, the data quality and consistency can be improved, the abnormal state in motor operation can be effectively identified, data support is provided for motor performance evaluation and fault early warning, and the method is suitable for real-time monitoring and operation optimization of a motor system.

Description

Technical Field

[0001] The present invention belongs to the technical field of power electronics, and in particular relates to a motor performance evaluation method based on multi-parameter fusion. Background Art

[0002] In the actual operation of a motor system, its performance is affected by a combination of electrical, mechanical, thermal and other multi-dimensional parameters. Monitoring and analysis of a single parameter can no longer fully and accurately assess the real-time operating status and potential performance risks of the motor. Traditional motor performance evaluation methods usually only perform independent analysis on a single type of parameter (such as current, vibration amplitude or temperature), lacking deep fusion and correlation mining of multi-source heterogeneous data. This leads to one-sided evaluation results and makes it difficult to effectively capture the coupling relationship between parameters and the performance evolution laws under complex working conditions. In addition, with the development of industrial intelligence and the complexity of motor systems, the multi-dimensional data generated during motor operation exhibits characteristics such as high noise, high dimensionality and strong time series. Traditional data processing and dimensionality reduction methods (such as simple statistical analysis or principal component analysis) are difficult to fully extract the core features that characterize motor performance. In addition, in anomaly detection scenarios, there is a lack of dynamic correlation analysis of parameter trend changes, fluctuation amplitude and threshold adaptability, which can easily lead to missed detection or misjudgment of anomalies. Summary of the Invention

[0003] Based on this, it is necessary to address the above technical issues and provide a multi-parameter fusion motor performance evaluation method that can achieve accurate evaluation of motor performance and reliable identification of abnormal conditions, so as to meet the high reliability and high stability operation requirements of motor systems in industrial scenarios.

[0004] In a first aspect, the present application provides a motor performance evaluation method based on multi-parameter fusion, including:

[0005] A multi-dimensional original data set including electrical signals, mechanical vibration signals and thermal temperature data is obtained during the operation of the motor, and the original data set is preprocessed and standardized to obtain a first data set.

[0006] The electrical, mechanical and thermal parameters in the first data set are reduced in dimension to extract the core parameter combination that affects the motor performance and generate the second feature data set.

[0007] The classification algorithm is used to perform classification and labeling based on the second feature data set, determine the distribution range of the key performance indicators, and obtain a third classification result set.

[0008] The third classification result set is used to build an anomaly detection model based on time series, analyze the changing trend of key performance indicators in different time periods, determine whether there are abnormal points that deviate from the preset threshold, and generate the fourth anomaly detection result.

[0009] In one embodiment, preprocessing and normalizing the original data set to obtain a first data set includes:

[0010] The voltage fluctuation waveform data of the electrical signal is obtained; the voltage fluctuation waveform data includes three-phase current and voltage sampling points of multiple cycles.

[0011] The mechanical vibration signal is subjected to time-frequency transformation to obtain the time-frequency distribution matrix of the vibration signal.

[0012] A temperature gradient field is generated according to the spatial distribution characteristics of thermal temperature data; the temperature gradient field is composed of the temperature change rates of multiple sensor nodes.

[0013] The time-frequency distribution matrix and the temperature gradient field are spatially and temporally aligned to generate a tensor structure that integrates multidimensional features.

[0014] The tensor structure is normalized at the channel level using preset normalization parameters, and a multi-scale feature set is extracted using dilated convolution with increasing kernel size. The normalization parameters are dynamically adjusted based on the peak-to-valley difference of the voltage fluctuation waveform data.

[0015] Each feature in the extracted multi-scale feature set is connected across layers to generate a fused feature vector with temporal correlation.

[0016] According to the dimensional matching result of the fused feature vector, the weight parameters of the fully connected layer are updated and adjusted through the back propagation algorithm to generate the first data set.

[0017] In one embodiment, a channel-level normalization is performed on the tensor structure using a preset normalization parameter, and a dilated convolution with increasing kernel size is used to extract a multi-scale feature set, including:

[0018] The tensor structure is normalized at the channel level according to the preset value of the normalization parameter to obtain a normalized tensor.

[0019] A convolution kernel size sequence is used to generate multiple dilated convolution kernel groups; the dilated convolution kernel groups contain increasing convolution kernel sizes and corresponding dilation rate parameters.

[0020] The standardized tensor is input into the dilated convolution kernel group for multi-scale feature extraction to generate a multi-scale feature set.

[0021] In one embodiment, dimensionality reduction is performed on the electrical, mechanical, and thermal parameters in the first data set to extract the core parameter combinations that affect the motor performance and generate a second feature data set, including:

[0022] The electrical parameters, mechanical parameters, and thermal parameters of the first data set are obtained; the parameters include voltage fluctuation rate, bearing vibration amplitude, and winding temperature rise gradient.

[0023] The covariance matrix between the parameters is calculated, and principal component analysis is performed based on the covariance matrix to extract parameter combinations whose principal component contribution rates exceed a preset threshold; the parameter combinations include rotor eccentricity and heat dissipation efficiency coefficient.

[0024] The characteristic orthogonality of the parameter combinations was evaluated and the principal component loading matrix was generated.

[0025] The core parameter combination is determined according to the principal component load matrix and parameter redundancy screening rules; the core parameter combination includes the current harmonic distortion rate and the thermal conduction attenuation coefficient.

[0026] The core parameter combination and the parameter sensitivity sequence are multi-dimensionally mapped to generate the second feature data set.

[0027] In one embodiment, the principal component loading matrix is ​​constructed using the following formula:

[0028]

[0029] Where L represents the principal component loading matrix, U k represents the first k left singular vector matrices of X, Λ k =diag(λ1,λ2,...λ k ) represents the diagonal matrix of the first k eigenvalues, and X represents the normalized parameter matrix.

[0030] In one embodiment, a classification algorithm is used to perform classification and labeling based on the second feature data set to determine the distribution range of the key performance indicators, thereby obtaining a third classification result set, including:

[0031] The efficiency change critical value parameter and the vibration abnormality feature point vector in the second feature data set are obtained.

[0032] The efficiency change critical value parameter is input into the pre-trained classification model to obtain the critical value classification label.

[0033] A multi-dimensional feature matrix is ​​extracted based on the vibration abnormality feature point vector, and a clustering algorithm is used to generate abnormal feature classification labels.

[0034] Cross-match the critical value classification labels with the abnormal feature classification labels to determine the joint distribution interval of the key performance indicators.

[0035] The second feature data set is dynamically segmented based on the joint distribution interval to generate a third classification result set containing annotation information.

[0036] The third classification result set includes a data subset corresponding to the joint distribution interval and associated classification label parameters.

[0037] In one embodiment, the third classification result set is used to build a time series-based anomaly detection model to analyze the changing trends of key performance indicators in different time periods, including:

[0038] Get the time window division parameters of the third classification result set; the time window division parameters include the start point and end point of different time periods.

[0039] The key performance indicators are segmented and intercepted according to the time window partition parameters to obtain indicator subsequences with timestamps.

[0040] The statistical feature vector of the indicator subsequence is extracted and input into the anomaly detection model, and the adaptive threshold interval and indicator change trend are generated according to the historical data distribution.

[0041] In one embodiment, after generating the adaptive threshold interval, the method further includes:

[0042] Determine whether the trend slope of the indicator change trend exceeds the upper or lower limit of the adaptive threshold interval. If it exceeds, mark it as a potential anomaly point.

[0043] Calculate the deviation of potential outliers; the deviation is the absolute value of the difference between the actual slope and the boundary of the adaptive threshold interval.

[0044] The deviation is weighted and corrected in combination with the fluctuation amplitude to generate an abnormality confidence score.

[0045] The anomaly confidence score is calculated using the following formula:

[0046]

[0047]

[0048] Among them, Confidencce t represents the anomaly confidence score, b t Represents the threshold boundary (L t or U t ), σ s represents the standard deviation of the historical slope, σ υ Indicates the standard deviation of historical fluctuations, υ t represents the fluctuation amplitude, t represents the potential abnormal point, s t It represents the trend slope of the indicator sequence, α represents the influence strength of controlling the slope deviation, β represents the nonlinear enhancement coefficient, and γ represents the weight of the fluctuation amplitude.

[0049] The potential abnormal points are sorted according to the abnormality confidence scores, and the points exceeding the preset score threshold are screened out as the fourth abnormality detection results.

[0050] In a second aspect, the present application further provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the above method when executing the computer program.

[0051] In a third aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the above method when executed by a processor.

[0052] The above-mentioned multi-parameter fusion motor performance evaluation method, computer equipment and storage medium first collect electrical signals, mechanical vibration signals and thermal temperature data during motor operation, and form a first data set through preprocessing and standardization; secondly, the multi-dimensional parameters of the first data set are reduced in dimension, and the core parameter combination is extracted to generate a second feature data set; then, the second feature data set is classified and labeled using a classification algorithm, the distribution range of key performance indicators is determined, and a third classification result set is obtained; finally, a time series anomaly detection model is constructed based on the third classification result set to analyze the indicator change trend and identify anomalies that deviate from the preset threshold, and generate a fourth anomaly detection result. This method improves data quality and consistency through multi-dimensional data preprocessing and standardization; uses dimensionality reduction technology to extract core parameters, reduce computational complexity and retain key information; classification labeling and time series modeling realize performance indicator distribution analysis and abnormal trend detection, which can effectively identify abnormal conditions during motor operation, provide data support for motor performance evaluation and fault warning, enhance the accuracy of evaluation results and detection efficiency, and is suitable for real-time monitoring and operation optimization of motor systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 A flowchart of a motor performance evaluation method using multi-parameter fusion provided by an embodiment of the present invention;

[0055] Figure 2 A flowchart of an embodiment of the present invention for performing dimensionality reduction processing on electrical, mechanical, and thermal parameters in a first data set, extracting core parameter combinations that affect motor performance, and generating a second feature data set;

[0056] Figure 3 The embodiment of the present invention provides a flowchart for performing classification and labeling based on the second feature data set using a classification algorithm, determining the distribution range of key performance indicators, and obtaining a third classification result set. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0058] In one embodiment, Figure 1 As shown, the present application provides a motor performance evaluation method based on multi-parameter fusion, which may include the following steps:

[0059] Step S101 : obtaining a multi-dimensional original data set including electrical signals, mechanical vibration signals, and thermal temperature data during the operation of the motor, and preprocessing and standardizing the original data set to obtain a first data set.

[0060] Specifically, electrical signals (such as three-phase voltage and current waveform sampling points), mechanical vibration signals (time-domain vibration amplitude sequence), and thermal temperature data (multi-sensor node temperature values) are acquired during motor operation to form a heterogeneous raw data set with time domain, frequency domain, and spatial distribution. The first data set is generated through the following processing:

[0061] Electrical signal processing: Perform period segmentation and noise filtering on the voltage fluctuation waveform to extract statistical features such as voltage fluctuation rate and current harmonic distortion rate.

[0062] Mechanical vibration processing: The vibration signal is converted into a time-frequency distribution matrix through short-time Fourier transform (STFT) to capture the energy distribution of different frequency components.

[0063] Thermal data processing: Calculate the temperature gradient field based on the spatial layout of sensors and quantify the temperature change rate between adjacent nodes.

[0064] Data normalization: Unify multimodal data into a tensor structure (such as time × feature × channel), and use dynamic normalization parameters (such as adjusting the scaling factor based on the voltage peak-to-valley difference) to perform channel-level normalization to eliminate dimensional differences.

[0065] Step S102 , performing dimensionality reduction processing on the electrical, mechanical and thermal parameters in the first data set, extracting the core parameter combination that affects the motor performance, and generating a second feature data set.

[0066] Specifically, for the electrical parameters (such as voltage fluctuation rate), mechanical parameters (such as bearing vibration amplitude) and thermal parameters (such as winding temperature rise gradient) in the first data set, the covariance matrix between the parameters is calculated, and principal component analysis is performed based on the matrix to extract parameter combinations whose principal component contribution rates exceed the preset threshold (such as rotor eccentricity and heat dissipation efficiency coefficient). The characteristic orthogonality is evaluated through the principal component load matrix (constructed by the product of the right singular vector and the square root of the eigenvalue). The redundant parameters are eliminated in combination with the parameter redundancy screening rules, and the core parameter combination including the current harmonic distortion rate and the thermal conduction attenuation coefficient is determined. It is then multi-dimensionally mapped with the parameter sensitivity sequence to generate a second characteristic data set focusing on key influencing factors.

[0067] Step S103 , classifying and labeling the second feature data set using a classification algorithm, determining the distribution range of the key performance indicators, and obtaining a third classification result set.

[0068] Furthermore, based on the second feature data set, the efficiency change critical value parameters therein are input into a pre-trained classification model to obtain the critical value classification label, the multi-dimensional feature matrix is ​​extracted from the vibration abnormality feature point vector and the clustering algorithm is used to generate the abnormal feature classification label, and the joint distribution interval of the key performance indicators (such as normal, warning, and fault intervals) is determined by cross-matching the two types of labels. Based on the interval, the second feature data set is dynamically segmented to generate a third classification result set containing data subsets and associated classification label parameters, thereby realizing the performance status labeling of the core parameters.

[0069] Step S104: construct a time series-based anomaly detection model using the third classification result set, analyze the changing trend of the key performance indicators in different time periods, determine whether there are abnormal points that deviate from the preset threshold, and generate a fourth anomaly detection result.

[0070] Specifically, the time window division parameters of the third classification result set (including the starting and ending points of different time periods) are used to segment the key performance indicators to obtain indicator subsequences with timestamps. The statistical feature vectors of the subsequences are extracted and input into the anomaly detection model. Based on the historical data distribution, adaptive threshold intervals and indicator change trends are generated. Potential abnormal points are marked by judging whether the trend slope exceeds the upper and lower limits of the threshold interval. The deviation of the absolute value of the difference between the actual slope and the threshold boundary is calculated, and a weighted correction is performed based on the fluctuation amplitude to generate an abnormality confidence score. The potential abnormal points are sorted according to the score, and the points that exceed the preset score threshold are screened out to generate the fourth abnormality detection result, thereby realizing the identification and positioning of the temporal abnormalities of the motor operation status.

[0071] The above-mentioned multi-parameter fusion motor performance evaluation method first collects the electrical signals, mechanical vibration signals and thermal temperature data during motor operation, and forms a first data set through preprocessing and standardization; secondly, the multi-dimensional parameters of the first data set are reduced in dimensionality, and the core parameter combination is extracted to generate a second feature data set; then, the second feature data set is classified and labeled using a classification algorithm, the distribution range of key performance indicators is determined, and a third classification result set is obtained; finally, a time series anomaly detection model is constructed based on the third classification result set to analyze the indicator change trend and identify anomalies that deviate from the preset threshold, generating a fourth anomaly detection result. This method improves data quality and consistency through multi-dimensional data preprocessing and standardization; uses dimensionality reduction technology to extract core parameters, reduce computational complexity and retain key information; classification labeling and time series modeling realize performance indicator distribution analysis and abnormal trend detection, which can effectively identify abnormal conditions during motor operation, provide data support for motor performance evaluation and fault warning, enhance the accuracy of evaluation results and detection efficiency, and is suitable for real-time monitoring and operation optimization of motor systems.

[0072] In one embodiment, preprocessing and normalizing the original data set to obtain the first data set may include the following steps:

[0073] Step S201 , obtaining voltage fluctuation waveform data of an electrical signal; the voltage fluctuation waveform data includes a plurality of cycles of three-phase current and voltage sampling points.

[0074] Step S202 : performing time-frequency transformation on the mechanical vibration signal to obtain a time-frequency distribution matrix of the vibration signal.

[0075] Step S203 : generating a temperature gradient field according to the spatial distribution characteristics of the thermal temperature data; the temperature gradient field is composed of the temperature change rates of multiple sensor nodes.

[0076] Step S204 , aligning the time-frequency distribution matrix with the temperature gradient field in time and space to generate a tensor structure integrating multi-dimensional features.

[0077] In step S205 , the tensor structure is normalized at the channel level using a preset normalization parameter, and a multi-scale feature set is extracted using a dilated convolution with increasing convolution kernel size; the normalization parameter is dynamically adjusted based on the peak-to-valley difference of the voltage fluctuation waveform data.

[0078] Preferably, first, a normalization operation is performed on each channel of the tensor structure according to the preset normalization parameters. By subtracting the channel mean and dividing by the standard deviation, the scale differences caused by different physical dimensions (such as vibration amplitude and temperature value) are eliminated to make the data distribution uniform. Among them, the normalization parameter is dynamically adjusted based on the peak-to-valley difference of the voltage fluctuation waveform data, which can adapt to the fluctuation range of the data under different working conditions and improve the accuracy of standardization. Secondly, a hole convolution kernel group with increasing convolution kernel size is used to extract features from the standardized tensor. By setting convolution kernels of different sizes (such as 3, 5, 7) and hole rates (such as 1, 2, 4), the receptive field is expanded without increasing too much computational effort, and a multi-scale feature set from local details (such as high-frequency components of vibration signals) to global trends (such as temperature field change trends) is extracted in parallel. This process not only retains the characteristic information of the original data at different scales, but also enhances the adaptability of the model to complex working conditions.

[0079] Step S206 , cross-layer connection is performed on each feature in the extracted multi-scale feature set to generate a fused feature vector with temporal correlation.

[0080] Step S207 , updating and adjusting the weight parameters of the fully connected layer through the back propagation algorithm according to the dimensional matching result of the fused feature vector to generate a first data set.

[0081] Specifically, a standardized feature dataset is generated by collaboratively processing heterogeneous electrical, mechanical, and thermal data from multiple sources during motor operation. First, voltage fluctuation waveform data, including multi-cycle three-phase current and voltage sampling points, is acquired. Simultaneously, a time-frequency transformation is performed on the mechanical vibration signal to generate a time-frequency distribution matrix. Based on the spatial distribution characteristics of the thermal temperature data, a temperature gradient field, consisting of the temperature change rates of multiple sensor nodes, is constructed. The time-frequency distribution matrix and the temperature gradient field are spatially and temporally aligned to construct a tensor structure that fuses multidimensional features. This tensor structure is then channel-wise normalized using a normalization parameter dynamically adjusted based on the peak-to-valley difference of the voltage fluctuations. A multi-scale feature set is extracted using a dilated convolution with increasing kernel size. Multi-scale features are integrated through cross-layer connections to generate a fused feature vector with temporal correlation. Based on the dimensionality matching results, the weight parameters of the fully connected layer are optimized using a backpropagation algorithm, ultimately generating a standardized first dataset.

[0082] This embodiment effectively integrates multi-dimensional sensor data through spatiotemporal alignment and dynamic normalization technology, solving the problem of spatiotemporal scale differences between different physical quantities; adopts incremental hole convolution and cross-layer connection mechanism to enhance the correlation expression of time series features while retaining the multi-scale characteristics of the original data; through dynamic adjustment of normalization parameters and weight optimization, the model's adaptability to changes in data distribution under different working conditions is improved, providing high-quality standardized input for subsequent feature dimensionality reduction and anomaly detection, and enhancing the robustness and generalization ability of the motor performance evaluation model.

[0083] In one embodiment, channel-level normalization of the tensor structure is performed using a preset normalization parameter, and a dilated convolution with increasing convolution kernel size is used to extract a multi-scale feature set, which may include the following steps:

[0084] Step S301 : performing channel-level normalization processing on the tensor structure according to a preset value of a normalization parameter to obtain a normalized tensor.

[0085] Step S302: Generate multiple dilated convolution kernel groups using a convolution kernel size sequence; the dilated convolution kernel groups include increasing convolution kernel sizes and corresponding dilation rate parameters.

[0086] Step S303: Input the normalized tensor into the dilated convolution kernel group to perform multi-scale feature extraction to generate a multi-scale feature set.

[0087] Specifically, the tensor structure that integrates multi-dimensional features is subjected to channel-level normalization processing based on preset normalization parameters. By performing mean zeroing and variance normalization on the data of each channel, the differences in different physical dimensions are eliminated to obtain a standardized tensor. Subsequently, based on the convolution kernel size sequence, a dilated convolution kernel group containing increasing convolution kernel sizes (such as 3, 5, 7) and corresponding dilation rate parameters (such as 1, 2, 4) is constructed. The standardized tensor is input into this kernel group. Through convolution operations with different dilation rates, multi-scale local features and global dependencies are extracted in parallel, generating a multi-scale feature set covering different receptive fields.

[0088] This embodiment improves the compatibility of multi-source heterogeneous data by unifying data distribution, ensuring the stability of subsequent feature extraction. The dilated convolution kernel group, through a combination of increasing size and dilation ratio, expands the spatial scale range of feature extraction without significantly increasing computational complexity. This approach can capture both high-frequency local anomaly details in mechanical vibration signals and low-frequency global distribution trends in thermal temperature fields. The generation of multi-scale feature sets effectively preserves the feature details and cross-domain correlation information of different dimensions in the original data, providing a more comprehensive feature representation for subsequent dimensionality reduction and performance evaluation, enhancing the model's adaptability to complex working conditions and its feature characterization capabilities.

[0089] In one embodiment, Figure 2As shown, performing dimensionality reduction processing on the electrical, mechanical and thermal parameters in the first data set, extracting the core parameter combination that affects the motor performance, and generating the second feature data set may include the following steps:

[0090] Step S401: Acquire electrical parameters, mechanical parameters, and thermal parameters of a first data set; the parameters include voltage fluctuation rate, bearing vibration amplitude, and winding temperature rise gradient.

[0091] Step S402 , calculating the covariance matrix between parameters, performing principal component analysis based on the covariance matrix, and extracting parameter combinations whose principal component contribution rates exceed a preset threshold; the parameter combinations include rotor eccentricity and heat dissipation efficiency coefficient.

[0092] Preferably, after obtaining the multi-dimensional parameters of the first data set, high-dimensional data dimensionality reduction is achieved through covariance matrix and principal component analysis (PCA). First, the covariance matrix between electrical, mechanical, and thermal parameters is calculated. Where Y represents the standardized parameter matrix (n represents the number of samples, rows represent samples, and columns represent parameters such as voltage fluctuation rate, bearing vibration amplitude, etc.). The element ∑ij of the covariance matrix reflects the linear correlation between the i-th parameter and the j-th parameter. A positive value indicates a change in the same trend, and a negative value indicates a change in the opposite trend. The larger the absolute value, the stronger the correlation. The core of principal component analysis based on the covariance matrix is ​​to use eigenvalue decomposition ∑=VΛV T Extract the principal components. Where V represents the eigenvector matrix, the column vector corresponds to the principal component direction; Λ represents the eigenvalue diagonal matrix, the eigenvalue Indicates the data variance explained by the corresponding principal component. The larger the value, the more important the information contained in the principal component. By calculating the cumulative contribution rate (p represents the total number of parameters), and the top k principal components whose cumulative contribution rate exceeds a preset threshold (such as 85%) are screened out. The corresponding parameter combination (such as rotor eccentricity and heat dissipation efficiency coefficient) is the comprehensive feature that has the greatest impact on motor performance.

[0093] Step S403: perform characteristic orthogonality evaluation on the parameter combination to generate a principal component loading matrix.

[0094] Step S404 , determining a core parameter combination according to the principal component load matrix and parameter redundancy screening rules; the core parameter combination includes current harmonic distortion rate and heat conduction attenuation coefficient.

[0095] Step S405 : Perform multi-dimensional mapping on the core parameter combination and the parameter sensitivity sequence to generate a second feature data set.

[0096] Specifically, electrical, mechanical, and thermal parameters, including voltage fluctuation rate, bearing vibration amplitude, and winding temperature rise gradient, are extracted from the first data set. The correlation is quantified by calculating the covariance matrix between the parameters. Based on principal component analysis (PCA), a linear transformation is performed on the high-dimensional parameter space to extract parameter combinations whose principal component contribution rate exceeds a preset threshold (e.g., 85%), such as rotor eccentricity and heat dissipation efficiency coefficient. The principal component load matrix is ​​generated through feature orthogonality evaluation. Combined with parameter redundancy screening rules (e.g., eliminating parameters with load absolute values ​​below a threshold), the core parameter combination, including current harmonic distortion rate and thermal conduction attenuation coefficient, is determined. This combination is then multi-dimensionally mapped with the parameter sensitivity sequence to generate a second feature data set focusing on key influencing factors.

[0097] This embodiment effectively identifies the linear coupling relationship between multiple parameters through covariance matrix and principal component analysis, compresses redundant information in a data-driven manner, reduces computational complexity while retaining the main factors affecting motor performance. The combination of the principal component load matrix and the redundancy screening rules ensures the orthogonality and physical interpretability of the core parameter combination and avoids information overlap between features. The multidimensional mapping mechanism associates the core parameters with the sensitivity sequence, enhancing the responsiveness of the feature set to changes in the motor's operating state, providing low-dimensional, high-discrimination input features for subsequent classification labeling and anomaly detection, and improving the training efficiency and generalization performance of the evaluation model.

[0098] In one embodiment, the principal component loading matrix can be constructed using the following formula:

[0099]

[0100] Where L represents the principal component loading matrix, U k represents the first k left singular vector matrices of X, Λ k =diag(λ1,λ2,...λ k ) represents the diagonal matrix of the first k eigenvalues, and X represents the normalized parameter matrix.

[0101] Preferably, Λ k =diag(λ1,λ2,...λ k ), λ i is the matrix X T The first k eigenvalues ​​of X (also known as the singular values ​​σ of X) i Calculated, ), and satisfy λ1≥λ2≥...≥λ k ≥0.

[0102] This embodiment uses singular value decomposition to efficiently extract key eigenvalues ​​from a high-dimensional parameter matrix. The principal components are normalized using an eigenvalue diagonal matrix, ensuring that the load matrix elements accurately reflect the correlation between the parameters and the principal components. The principal component load matrix provides a quantitative metric for core parameter screening. Combined with parameter redundancy rules, this effectively eliminates redundant parameters, reducing data dimensionality while retaining key factors influencing motor performance. This improves the computational efficiency and accuracy of subsequent performance evaluation and anomaly detection models.

[0103] In one embodiment, Figure 3 As shown, the following steps may be included:

[0104] Step S501 : obtaining efficiency change critical value parameters and vibration abnormality feature point vectors in a second feature data set.

[0105] Step S502 : Input the efficiency change critical value parameter into a pre-trained classification model to obtain a critical value classification label.

[0106] Preferably, after obtaining the efficiency change critical value parameters (such as the current and voltage parameters corresponding to the turning point of the motor load efficiency) in the second feature data set, it is necessary to use a pre-trained classification model to label their status. The classification model (such as support vector machine SVM, random forest, etc.) learns the mapping relationship between efficiency change and performance status through historical annotation data. The input layer receives the normalized critical value parameter vector, and the output layer outputs the classification label (such as "normal", "warning", "fault") after feature space transformation. This process maps continuous parameter values ​​to discrete performance status categories through a supervised learning mechanism, thereby achieving a quantitative evaluation of motor efficiency-related performance. For example, when the input parameter combination indicates that the efficiency is lower than 90% of the rated value and continues to decline, the model outputs a "warning" label, indicating that the motor may enter an inefficient operating range. The introduction of the classification model provides a data-driven objective classification standard for key performance indicators, avoids the subjectivity of manual threshold setting, and improves the consistency and reliability of the labeling results.

[0107] Step S503 : extracting a multi-dimensional feature matrix based on the vibration abnormality feature point vector, and generating abnormal feature classification labels using a clustering algorithm.

[0108] Step S504 : cross-matching the critical value classification labels with the abnormal feature classification labels to determine the joint distribution interval of the key performance indicators.

[0109] Step S505 : Dynamically segment the second feature data set based on the joint distribution interval to generate a third classification result set containing annotation information.

[0110] The third classification result set includes a data subset corresponding to the joint distribution interval and associated classification label parameters.

[0111] First, efficiency change threshold parameters and vibration anomaly feature point vectors are extracted from the second feature dataset. The former are fed into a pre-trained classification model to obtain threshold classification labels, while the latter are generated by extracting a multidimensional feature matrix and applying a clustering algorithm to generate anomaly feature classification labels. The two labels are then cross-matched to determine the joint distribution interval of the key performance indicators. Finally, the second feature dataset is dynamically segmented based on the joint distribution interval, generating a third classification result set containing data subsets and their corresponding classification label parameters.

[0112] This embodiment combines a classification model with a clustering algorithm to achieve multi-dimensional classification and precise labeling of motor performance indicators. The classification model processes efficiency change thresholds to determine performance status based on historical data learning. The clustering algorithm analyzes vibration anomaly characteristics and mines inherent patterns in the data. A cross-matching and dynamic segmentation mechanism effectively integrates classification results from different dimensions to form a joint distribution interval. This allows the third classification result set to reflect both the overall motor performance and detailed information on each key indicator, improving the accuracy and efficiency of anomaly identification.

[0113] In one embodiment, using the third classification result set to build a time series-based anomaly detection model to analyze the changing trends of key performance indicators in different time periods may include the following steps:

[0114] Step S601: Obtain time window division parameters of the third classification result set; the time window division parameters include the starting point and the ending point of different time periods.

[0115] Step S602 : segmenting the key performance indicators according to the time window partitioning parameters to obtain indicator subsequences with timestamps.

[0116] Step S603: extract the statistical feature vector of the indicator subsequence and input it into the anomaly detection model, and generate an adaptive threshold interval and indicator change trend according to the historical data distribution.

[0117] Specifically, the time window partitioning parameters from the third classification result set are first obtained. These parameters define the start and end points of different time periods. Based on this, the key performance indicators are segmented and intercepted to form indicator subsequences with timestamps to preserve the temporal characteristics of the data. Next, the statistical feature vectors (such as mean, variance, slope, etc.) of each indicator subsequence are extracted and input into the anomaly detection model. Based on the historical data distribution, the model dynamically generates adaptive threshold intervals and analyzes the indicator change trends to provide a judgment basis for anomaly detection.

[0118] This processing method, based on time windows and adaptive thresholds, effectively combines the timing characteristics of motor performance indicators with the data distribution characteristics. Time window division structures the data in the temporal dimension, facilitating the capture of performance changes over time. Statistical feature extraction focuses on key data information, reducing data dimensionality while retaining effective features. Adaptive threshold intervals can be dynamically adjusted based on historical data, overcoming the limitation of fixed thresholds in adapting to complex operating conditions. This improves the responsiveness of anomaly detection to changes in motor operating status, allowing for more accurate identification of potential anomalies and enhancing the reliability and real-time nature of motor performance monitoring.

[0119] In one embodiment, after the adaptive threshold interval is generated, the following steps may be further included:

[0120] Step S701: determine whether the trend slope of the indicator change trend exceeds the upper limit or lower limit of the adaptive threshold range. If exceeded, mark it as a potential abnormal point.

[0121] Step S702 , calculating the deviation of the potential outlier; the deviation is the absolute value of the difference between the actual slope and the boundary of the adaptive threshold interval.

[0122] Step S703: Perform weighted correction on the deviation in combination with the fluctuation amplitude to generate an abnormality confidence score.

[0123] The anomaly confidence score is calculated using the following formula:

[0124]

[0125] Among them, Confidencce t represents the anomaly confidence score, b t Represents the threshold boundary (L t or U t ), σ s represents the standard deviation of the historical slope, σ υ Indicates the standard deviation of historical fluctuations, υ t represents the fluctuation amplitude, t represents the potential abnormal point, s t It represents the trend slope of the indicator sequence, α represents the influence strength of controlling the slope deviation, β represents the nonlinear enhancement coefficient, and γ represents the weight of the fluctuation amplitude.

[0126] Step S704 , sorting the potential abnormal points according to the abnormality confidence scores, and selecting the points exceeding a preset score threshold as the fourth abnormality detection result.

[0127] After obtaining the adaptive threshold interval and indicator trend, potential outliers are identified by determining whether the indicator trend slope exceeds the upper and lower limits of the adaptive threshold interval. For each potential outlier, the absolute difference between its actual slope and the threshold interval boundary is calculated as the deviation. This is then weighted and corrected using a formula based on the fluctuation amplitude to generate an anomaly confidence score. Finally, the potential outliers are ranked according to the scores, and those exceeding the preset threshold are selected as the fourth anomaly detection result.

[0128] Preferably, when s t When the threshold is exceeded, Confidencce t With deviation |s t -b t | increases exponentially (the growth rate is controlled by β), and the fluctuation amplitude υ t The scores are further modified through index terms to achieve a joint evaluation of slope deviation and fluctuation amplitude.

[0129] When s t In the threshold range, Confidencce t =1, indicating that the point is normal.

[0130] By σ s and σ υ Normalization eliminates the dimensional differences of different parameters and makes the formula universal.

[0131] This anomaly detection method combines slope deviation and fluctuation amplitude to quantitatively assess potential anomalies. Adaptive threshold intervals improve the detection model's adaptability to changing motor operating conditions, avoiding misjudgments or missed detections caused by fixed thresholds. The anomaly confidence score formula comprehensively considers slope deviation and fluctuation characteristics, flexibly adjusting weights through parameters to ensure that detection results are more consistent with actual operating conditions. A sorting and screening mechanism ensures that high-confidence anomalies are identified first, providing a reliable basis for motor fault warnings and enhancing the accuracy and effectiveness of motor operating status monitoring.

[0132] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0133] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the motor performance evaluation method based on multi-parameter fusion as described above are implemented.

[0134] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0135] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0136] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A motor performance evaluation method based on multi-parameter fusion, characterized in that: The method comprises: Acquire a multi-dimensional original data set including electrical signals, mechanical vibration signals, and thermal temperature data during the operation of the motor, and preprocess and standardize the original data set to obtain a first data set; Performing dimensionality reduction processing on the electrical, mechanical, and thermal parameters in the first data set to extract core parameter combinations that affect motor performance and generate a second feature data set; Performing classification and labeling based on the second feature data set using a classification algorithm to determine the distribution range of key performance indicators to obtain a third classification result set; The third classification result set is used to construct an anomaly detection model based on time series, analyze the changing trend of the key performance indicators in different time periods, determine whether there are abnormal points that deviate from the preset threshold, and generate a fourth anomaly detection result.

2. The method according to claim 1, characterized in that The preprocessing and standardizing of the original data set to obtain a first data set includes: Acquiring voltage fluctuation waveform data of the electrical signal; the voltage fluctuation waveform data comprising a plurality of cycles of three-phase current and voltage sampling points; Performing time-frequency transformation on the mechanical vibration signal to obtain a time-frequency distribution matrix of the vibration signal; Generate a temperature gradient field based on the spatial distribution characteristics of the thermal temperature data; the temperature gradient field is composed of the temperature change rates of multiple sensor nodes; Performing spatial and temporal alignment of the time-frequency distribution matrix and the temperature gradient field to generate a tensor structure integrating multi-dimensional features; The tensor structure is normalized at the channel level using a preset normalization parameter, and a multi-scale feature set is extracted using a dilated convolution with increasing convolution kernel size; the normalization parameter is dynamically adjusted based on the peak-to-valley difference of the voltage fluctuation waveform data; Performing cross-layer connections on the features in the extracted multi-scale feature set to generate a fused feature vector with temporal correlation; The weight parameters of the fully connected layer are updated and adjusted through a back propagation algorithm according to the dimensional matching result of the fused feature vector to generate a first data set.

3. The method according to claim 2, characterized in that The method of using a preset normalization parameter to perform channel-level normalization on the tensor structure and using a dilated convolution with increasing convolution kernel size to extract a multi-scale feature set includes: Performing channel-level normalization processing on the tensor structure according to a preset value of the normalization parameter to obtain a normalized tensor; A plurality of dilated convolution kernel groups are generated using a convolution kernel size sequence; the dilated convolution kernel groups include increasing convolution kernel sizes and corresponding dilation rate parameters; The standardized tensor is input into the dilated convolution kernel group to perform multi-scale feature extraction to generate a multi-scale feature set.

4. The method according to claim 1, wherein The step of performing dimensionality reduction processing on the electrical, mechanical, and thermal parameters in the first data set, extracting a core parameter combination that affects motor performance, and generating a second feature data set includes: Acquiring electrical parameters, mechanical parameters, and thermal parameters of the first data set; the parameters include voltage fluctuation rate, bearing vibration amplitude, and winding temperature rise gradient; Calculating a covariance matrix between the parameters, performing principal component analysis based on the covariance matrix, and extracting a parameter combination whose principal component contribution rate exceeds a preset threshold; the parameter combination includes rotor eccentricity and heat dissipation efficiency coefficient; Performing characteristic orthogonality evaluation on the parameter combination to generate a principal component loading matrix; Determining a core parameter combination according to the principal component load matrix and parameter redundancy screening rules; the core parameter combination includes current harmonic distortion rate and heat conduction attenuation coefficient; The core parameter combination and the parameter sensitivity sequence are multi-dimensionally mapped to generate a second feature data set.

5. The method according to claim 4, characterized in that The principal component loading matrix is ​​constructed using the following formula: Where L represents the principal component loading matrix, U k represents the first k left singular vector matrices of X, Λ k =diag(λ1,λ2,...λ k ) represents the diagonal matrix of the first k eigenvalues, and X represents the normalized parameter matrix.

6. The method according to claim 1, characterized in that The method of performing classification and labeling based on the second feature data set using a classification algorithm to determine the distribution range of key performance indicators and obtain a third classification result set includes: Obtaining the efficiency change critical value parameter and the vibration abnormality feature point vector in the second feature data set; Inputting the efficiency change critical value parameter into a pre-trained classification model to obtain a critical value classification label; Extracting a multidimensional feature matrix based on the vibration abnormality feature point vector, and generating abnormal feature classification labels using a clustering algorithm; Cross-matching the critical value classification label with the abnormal feature classification label to determine the joint distribution interval of the key performance indicator; Dynamically segmenting the second feature data set based on the joint distribution interval to generate a third classification result set containing annotation information; The third classification result set includes a data subset corresponding to the joint distribution interval and associated classification label parameters.

7. The method according to claim 1, characterized in that The method of constructing a time series-based anomaly detection model using the third classification result set and analyzing the changing trends of the key performance indicators in different time periods includes: Obtaining time window division parameters for the third classification result set; the time window division parameters include starting points and ending points of different time periods; Segmentally intercepting the key performance indicators according to the time window division parameters to obtain indicator subsequences with timestamps; The statistical feature vector of the indicator subsequence is extracted and input into the anomaly detection model, and the adaptive threshold interval and indicator change trend are generated according to the historical data distribution.

8. The method according to claim 7, characterized in that After generating the adaptive threshold interval, the method further includes: Determine whether the trend slope of the indicator change trend exceeds the upper limit or lower limit of the adaptive threshold range, and if so, mark it as a potential abnormal point; Calculating the deviation of the potential outlier; the deviation is the absolute value of the difference between the actual slope and the boundary of the adaptive threshold interval; Performing weighted correction on the deviation in combination with the fluctuation amplitude to generate an anomaly confidence score; The anomaly confidence score is calculated using the following formula: Among them, Confidencce t represents the anomaly confidence score, b t Represents the threshold boundary (L t or U t ), σ s represents the standard deviation of the historical slope, σ v Indicates the standard deviation of historical fluctuations, v t represents the fluctuation amplitude, t represents the potential abnormal point, s t It represents the trend slope of the indicator series, α represents the influence strength of controlling the slope deviation, β represents the nonlinear enhancement coefficient, and γ represents the weight of the fluctuation amplitude; The potential abnormal points are sorted according to the abnormality confidence scores, and points exceeding a preset score threshold are screened out as the fourth abnormality detection result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images